Australia's cyber agency urges AI defenses as incidents rise 20%
Australia's signals agency responded to 1,500 incidents in 2025-26, up 20%, and said services may go offline while they are secured.
The Australian Signals Directorate, the country's signals intelligence agency, urged businesses and critical infrastructure operators to adopt advanced AI models for cyber defense in its annual cyber threat report, released on Tuesday.
The agency also told Australians to expect more basic and important online services to go offline unexpectedly, saying “system security must be prioritised over system availability.” Organizations that follow that advice may take systems down to secure them rather than keep them running.
ASD responded to 1,500 cybersecurity incidents in the 2025-26 year, up 20% from the previous year, according to the report. It notified businesses, governments, and critical infrastructure operators of more than 4,400 potential malicious cyber incidents, an increase of 80%.
The report said hostile foreign governments, criminals, and hackers are using AI to gather intelligence, disrupt critical infrastructure, and steal money, finding vulnerabilities and launching attacks faster than before. Government and private networks faced “persistent” breach attempts from state-based actors, and ASD said actors “in the Indo-Pacific” could be preparing for potential future conflicts, including “future intelligence collection or disruption activities.”
We need AI tools to help automate our defences to stay ahead of AI-enabled attackers.
— Abigail Bradshaw, director general, Australian Signals Directorate
Bradshaw said legacy technology, outdated systems and devices their makers no longer update, was among the biggest risks and most attractive targets for hackers. The agency said cybersecurity was now a “core business risk and leadership responsibility” for all organizations. Defence Minister Richard Marles said Australians “must enhance their cyber defences while harnessing AI’s benefit.”
Bradshaw has previously spoken about ASD having access to Anthropic's Mythos model, which Anthropic said it could not release publicly because it could identify unknown flaws in IT systems. Labs have started giving defenders early access to such systems: Google DeepMind released Gemini 4 Carbon to cyber defenders first on October 28. In the week before the report, newly discovered vulnerabilities in PHP, the programming language that underpins most websites, set off a rush by governments and businesses to fix them.